Field measurements of PM_2.5 infiltration factor and portable air cleaner effectiveness during wildfire episodes in US residences

Abstract

Wildfires have frequently occurred in the western United States (US) during the summer and fall seasons in recent years. This study measures the PM_2.5 infiltration factor in seven residences recruited from five dense communities in Seattle, Washington, during a 2020 wildfire episode and evaluates the impacts of HEPA-based portable air cleaner (PAC) use on reducing indoor PM_2.5 levels. All residences with windows closed went through an 18-to-24-hour no filtration session, with five of seven following that period with an 18-to-24-hour filtration session. Auto-mode PACs, which automatically adjust the fan speed based on the surrounding PM_2.5 levels, were used for the filtration session. 10-second resolved indoor PM_2.5 levels were measured in each residence’s living room, while hourly outdoor levels were collected from the nearest governmental air quality monitoring station to each residence. Additionally, a time-activity diary in minute resolution was collected from each household. With the impacts of indoor sources excluded, indoor PM_2.5 mass balance models were developed to estimate the PM_2.5 indoor/outdoor (I/O) ratios, PAC effectiveness, and decay-related parameters. Among the seven residences, the mean infiltration factor ranged from 0.33 (standard deviation [SD]: 0.06) to 0.76 (SD: 0.05). The use of auto-mode PAC led to a 48%−78% decrease of indoor PM_2.5 levels after adjusting for outdoor PM_2.5 levels and indoor sources. The mean (SD) air exchange rates ranged from 0.30 (0.13) h⁻¹ to 1.41 (3.18) h⁻¹ while the PM_2.5 deposition rate ranged from 0.10 (0.54) h⁻¹ to 0.49 (0.47) h⁻¹. These findings suggest that staying indoors, a common protective measure during wildfire episodes, is insufficient to prevent people’s excess exposure to wildfire smoke, and provides quantitative evidence to support the utilization of auto-mode PACs during wildfire events in the US.

1. Introduction

Wildfires have frequently occurred in the western United States (US) during the summer and fall seasons in recent years (Westerling, 2016). In 2020, a series of major wildfires were reported in the western US, including California (CA), Oregon (OR), and Washington State (WA), with more than 33000 km² of land burned and 13000 buildings destroyed (National Interagency Fire Center, 2020). According to the Intergovernmental Panel on Climate Change (IPCC) and other relevant studies, climate change will increase the length of wildfire seasons in the US and worsen air quality during wildfire seasons in the coming decades (Abatzoglou and Williams, 2016; Littell et al., 2009; Parry et al., 2007).

Wildfires smoke contains a wide range of hazardous air pollutants, including fine particulate matter (PM_2.5; particles with an aerodynamic diameter smaller than 2.5 μm) (Naeher et al., 2007; Urbanski et al., 2008). Acute exposure to wildfire-related PM_2.5 has been linked with cardiorespiratory morbidity and mortality (Arriagada et al., 2019; DeFlorio-Barker et al., 2019; Doubleday et al., 2020; Hutchinson et al., 2018; Johnston et al., 2011; Reid et al., 2019). Evidence has shown that air pollutants from wildfires can be transported several hundred to thousands of kilometers away (Guan et al., 2020; Kollanus et al., 2016; Moeltner et al., 2013; Munoz-Alpizar et al., 2017). Thus, exposure to wildfire-related hazardous air pollutants is of concern in many regions. During wildfire episodes, the public has been advised to stay indoors to reduce wildfire-related PM_2.5 exposure (US Environmental Protection Agency, 2019; Washington State Department of Health, 2015). Regarding the infiltration of outdoor PM_2.5, the question remains whether staying indoors is sufficiently effective to reduce the excess wildfire-related exposure. A previous study showed that the mean PM_2.5 infiltration factor in WA residences was ~0.7 for non-wildfire scenarios, indicating a large indoor/outdoor concentration ratio without accounting for indoor sources (Allen et al., 2003). Considering the differences in particle size distributions and compositions, as well as people’s behavior between wildfire- and non-wildfire-related scenarios (Matz et al., 2020), the PM_2.5 infiltration factor for residences during wildfire events remains unclear.

In addition to sheltering indoors, people are also recommended to use a portable air cleaner (PAC) during wildfire episodes. Studies have shown that PACs equipped with a high-efficiency particulate air (HEPA) filter can reduce indoor PM_2.5 concentration to varying degrees in non-wildfire scenarios (Cox et al., 2018; Cui et al., 2018; Huang et al., 2020; Kajbafzadeh et al., 2015; Morishita et al., 2018; Sharma and Balasubramanian, 2017; Tran et al., 2020). However, few studies have evaluated the efficacy of HEPA-based PACs during wildfire episodes (Barn et al., 2008; Stauffer et al., 2020). Barn et al. (2008) examined the efficacy of HEPA-based PACs in 17 residences in British Columbia, Canada. Stauffer et al. (2020) assessed the efficacy of a HEPA-based PAC in an office in Montana, US, during the 2018 wildfire episode. Additionally, Henderson et al. (2005) evaluated the efficacy of an electrostatic precipitator (ESP) air cleaner in two residences in Colorado, US, during the 2002 fire season, despite the concern of ozone generation (Xiang et al., 2016). Generally, there is still very limited evidence of HEPA-based PACs’ effectiveness for residential use during wildfire episodes, especially in the US.

Among all the working modes of commercially-available PACs, auto-mode distinguishes itself by automatically adjusting the fan speed based on the PM_2.5 levels measured by an integrated PM_2.5 sensor. This auto-mode PAC feature is widely used in residential environments since it is easy to operate and potentially energy-efficient. Additionally, a previous study conducted in WA has demonstrated that PACs run in auto-mode can be more effective than other operating modes in reducing residential PM_2.5 during non-wildfire scenarios (Huang et al., 2020). However, it remains unclear whether the PACs with auto-mode features stay effective during wildfire scenarios.

While calculating the infiltration factor and PAC effectiveness, Henderson et al. (2005) ignored indoor-generated particles and assumed a constant PM_2.5 deposition rate (0.2 h⁻¹) throughout the study, which may lead to a large bias. Barn et al. (2008) used a censoring algorithm to identify periods with significant indoor PM_2.5 sources. Specifically, they assumed the presence of indoor PM_2.5 sources if indoor levels increased by > 50% compared to those in the previous 30 minutes, while outdoor levels stayed relatively stable. However, this censoring algorithm, which may work for major indoor sources, cannot fully adjust for the impacts of indoor-generated PM_2.5. For example, they found PM_2.5 infiltration factors exceeding 1.0 for some days after applying the censoring algorithm. On the other hand, neither of the above studies separated indoor PM_2.5 decay from ventilation, deposition, and PAC filtration.

Using the natural experiment of the 2020 wildfire episode in the western US, this study aims to 1) measure PM_2.5 infiltration factor in residences during wildfire episodes; 2) evaluate the impact of HEPA-based auto-mode PAC use on indoor PM_2.5 levels; 3) develop an algorithm separating indoor PM_2.5 decay from ventilation, deposition, and PAC filtration.

2. Methods

2.1. Site and participants

Seven residences from five dense communities in Seattle, WA, were recruited based on the following eligibility criteria: 1) the building was naturally ventilated, and 2) there was a governmental air quality monitoring (AQM) station within 10 km of the building. Residences belonged to student and staff at the University of Washington who volunteered for the study. All residences’ building characteristics were collected, including housing type, size, built year, and ventilation type.

2.2. Experimental design

The study was conducted from September 16–18, 2020, during a wildfire episode in the western US. Fig. 1 depicts the study design, where the seven residences are denoted R1–R7. The windows and doors in all recruited residences were kept closed during the experiment period. Residences R1–R5 first went through an 18-to-24-hour no filtration session, followed by an 18-to-24-hour filtration session. A HEPA-based PAC (Air Purifier 2000i, Philips, Andover, US) was utilized in the filtration. This manufacturer specifies a clean air delivery rate (CADR) of 198 m³/h for dust and 179 m³/h for smoke for this PAC. The PACs were set in auto mode during the filtration session. Additionally, Residences R6–R7 only went through an 18-to-24-hour non-filtration session due to limited resources.

Fig. 1.

Study flow.

Real-time PM_2.5 monitors (see Appendix Fig. A1 for the profile and structure) were used to measure indoor PM_2.5 mass concentration, temperature, and relative humidity (RH) at 10-second intervals in each residence’s living room. This PM_2.5 monitor and its relevant models were developed by our group, and utilized in many prior studies (Gao et al., 2015; Huang et al., 2020; Ong et al., 2019; Stampfer et al., 2020). The monitor is equipped with an optical particle sensor (Plantower PMSA003, Beijing Ereach Technology, China). This particle sensor has been well validated in previous studies (Huang et al., 2020; Levy Zamora et al., 2018; Zusman et al., 2020) for measuring both ambient and residential PM_2.5. The PM_2.5 monitors were calibrated in a residential environment before the experiment (see more details in the Appendix).

During the experiment period, the primary active wildfire regions were hundreds of kilometers away from Seattle, either in the east (central WA) or south (OR and CA). The predominantly east/south wind transported wildfire-related PM_2.5 from these regions to Seattle (Washington State Department of Ecology, 2020). During the experiment, the hourly outdoor PM_2.5 concentrations were obtained from the nearest governmental AQM station to each residence (Washington State Department of Ecology, 2020). The recruited residences were on the downwind side of the AQM stations during the experiment session, with a distance of 2–10 km.

The PACs and PM_2.5 monitors were placed at the center of the living room in each residence, at least 1 meter away from each other and the walls. Participants were asked to report their household’s indoor activities at the first opportunity in minute resolution. The indoor activities include but are not limited to cooking (with methods specified), smoking, cleaning, candle burning, turning PAC on/off, and window opening/closing.

2.3. Indoor PM_2.5 mass balance model

Based on the dynamic mass balance model with indoor sources absent, indoor PM_2.5 concentrations in naturally ventilated residences obey Equation (1):

$$\frac{d{C}_{in}(t)}{dt}=p\cdot AER\cdot C_{out}(t)-\left(AER+k_d+k_{PAC}\right)\cdot C_{in}(t)$$ (1)

where C_in(t) and C_out(t) are indoor and outdoor PM_2.5 levels at time t, μg/m³, respectively; p is the PM_2.5 penetration factor (unitless); AER is the air exchange rate, h⁻¹; k_d is the rate constant for PM_2.5 deposition to indoor surfaces, h⁻¹; k_PAC is the indoor PM_2.5 decay rate from PAC filtration, h⁻¹.

Under steady-state conditions, dC_in(t)/dt = 0, the indoor PM_2.5 levels can be calculated as

$$C_{in}(t)=\frac{p\cdot AER}{AER+k_d+k_{PAC}}\cdot C_{out}(t)$$ (2)

Under such conditions, PM_2.5 indoor/outdoor (I/O) ratios can be calculated as

$$I/O(t)=\frac{C_{in}(t)}{C_{out}(t)}=\frac{p\cdot AER}{AER+k_d+k_{PAC}}$$ (3)

The I/O ratios, when the PACs are absent and present, are expressed as I/O_INF (PM_2.5 infiltration factor) and I/O_PAC, and shown in Equations (4) and (5), respectively.

$$I/O_{INF}=\frac{p\cdot AER}{AER+k_d}$$ (4)

$$I/O_{PAC}=\frac{p\cdot AER}{AER+k_d+k_{PAC}}$$ (5)

The effectiveness of a PAC, ε, is defined as (Zhang et al., 2011)

$$\epsilon =\frac{k_{PAC}}{AER+k_d+k_{PAC}}\times 100\%$$ (6)

Based on Equations (4)–(6), ε can be calculated as

$$\epsilon =\left(1-\frac{I/O_{PAC}}{I/O_{INF}}\right)\times 100\%$$ (7)

When indoor PM_2.5 levels are not in steady-state, C_in(t) can be calculated using a dynamic model as shown in Equation (8) by assuming p, AER, k_d, k_PAC, and C_out(t) are constants within a short period, such as one hour (Sun et al., 2019).

$$\begin{gathered} C_{in}(t)=\frac{p\cdot AER}{AER+k_d+k_{PAC}}\cdot C_{out}(t)+ \\ \left(C_{in}(0)-\frac{p\cdot AER}{AER+k_d+k_{PAC}}\cdot C_{out}(t)\right)\times e^{-\left(AER+k_d+k_{PAC}\right)\cdot t}=a+b\cdot e^{-k\cdot t} \end{gathered}$$ (8)

where C_in(0) is the initial indoor PM_2.5 level at time 0, μg/m³; a, b, and k are coefficients of the model.

Hence, the indoor PM_2.5 decay/increase curve can be fitted using an exponential model. As a result, the total decay rate (AER + k_d + k_PAC) can be estimated as the coefficient, k, in the model. The total decay rate k, with and without PAC filtration, are expressed as k_T and k_TN, respectively.

Previous studies have suggested that the PM_2.5 penetration factor (p) is relatively stable (Allen et al., 2003; Diapouli et al., 2013; Fisk and Chan, 2017a). Based on Fisk and Chan (2017a), p with windows closed was assumed to follow a cropped normal distribution with values between 0 and 1, with a mean (SD) of 0.97 (0.06). Therefore, AER, k_d, and k_PAC can be determined based on Equations (4), (5), and (8).

2.4. Data analysis

To maximize the wildfire-related PM_2.5 signal and minimize the impacts of indoor-generated PM_2.5, I/O_INF and I/O_PAC were calculated based on measurements made during periods meeting the following criteria: 1) close to steady-state indoor PM_2.5 concentrations; 2) absence of indoor-generated PM_2.5 impacts; and 3) hourly outdoor PM_2.5 levels that were ≥ 50 μg/m³. In contrast, outdoor PM_2.5 levels are generally ≤ 20 μg/m³ during non-wildfire periods (Washington State Department of Ecology, 2020; Xiang et al., 2020). For these analyses, the 10-second PM_2.5 concentrations were aggregated into hourly means.

Due to the altered conditions of the indoor source (e.g., cooking) and sink (e.g., PAC use), there were periods with continuously decreasing/increasing indoor PM_2.5 levels in some residences. The decay/increase curves were fitted for each residence during periods meeting the following criteria: 1) ≥ 10 minutes after the emissions (e.g., cooking) stopped; 2) ≥ 10 minutes after the PAC was turned on/off; 3) no kitchen range hood was in use; and 4) the curve was visually smooth. The fitting was conducted for each hour separately, assuming AER, k_d, k_PAC, and C_out(t) were unchanged in that hour. For these analyses, the 10-second PM_2.5 concentrations were aggregated into means per minute. The hourly results for total decay rate k_T and k_TN, with and without PAC filtration, were aggregated into residence-specific data separately based on Monte Carlo simulations (Xiang et al., 2019b). Specifically, k_T and k_TN were assumed to be lognormally distributed as AER, part of k_T and k_TN, followed a lognormal distribution based on previous studies (Allen et al., 2003; Persily et al., 2010). Means and standard deviations (SDs) for the lognormal distributions (i.e., meanlog and sdlog) of k_T and k_TN were calculated based on the calculated hourly means and SDs from the exponential fitting (Johnson et al., 1994) (see more details in the Appendix). 5000 draws were then generated from the lognormal distribution of k_T and k_TN for each hour. The overall mean and SD were calculated based on the combined simulated datasets for all hours in each residence.

Monte Carlo simulations were also performed to account for the uncertainties of other parameters. I/O_INF and I/O_PAC were assumed to follow lognormal distributions based on a previous study (Huang et al., 2020). The parameters for the lognormal distributions were calculated using the approach mentioned above. 5000 draws were then generated from the lognormal distribution of I/O_INF, I/O_PAC, k_T, and k_TN, and the normal distribution of p (Fisk and Chan, 2017a) for each residence to calculate ε, AER, k_d, and k_PAC. Means and SDs were then calculated for these parameters based on the simulations.

All calculations were done in R Version 3.3.0 (R Core Team, 2013), embedded in RStudio Version 1.1.456.

3. Results

3.1. Residence characteristics and time activities

Table 1 summarizes the enrolled residences’ housing characteristics and time activities during the experiment. There were six apartments and one house among the seven residences, ranging from 54 to 177 m² and built years ranging from 1906 to 2019. All seven residences kept windows and doors closed during the experiment. Cooking events were reported in all residences except R7, while no smoking, cleaning, or candle burning activities were reported in any residence. As mentioned above, Residences R1–R5 utilized the auto-mode PACs, while R6–R7 had no PACs in use.

Table 1.

Residence characteristics and time activities.

Residence	Housing type	House size (m2)	Built year	Periods with PAC on	Periods of cooking
R1	Apartment	54	1988	09/17 11:10 – 09/18 12:00	09/16 13:07 – 09/16 13:50, 09/17 20:10 – 09/17 21:30
R2	Apartment	76	1948	09/17 15:31 – 09/18 15:52	09/16 16:10 – 09/16 16:30, 09/17 16:00 – 09/17 17:00, 09/18 11:00 – 09/18 11:30
R3	House	177	1947	09/17 15:20 – 09/18 15:34	09/17 17:45 – 09/17 18:15
R4	Apartment	39	2018	09/17 17:02 – 09/18 17:42	09/16 19:22 – 09/16 19:44, 09/17 19:30 – 09/17 19:50
R5	Apartment	56	2019	09/18 10:21 – 09/19 14:40	09/17 17:55 – 09/17 18:23, 09/18 08:55 – 09/18 09:13
R6	Apartment	74	1906	-	09/17 20:51 – 09/17 21:18, 09/18 21:26 – 09/18 22:01
R7	Apartment	69	1988	-	-

3.2. Concentrations

Fig. 2 shows the one-minute resolved mean outdoor and indoor PM_2.5 concentrations for each residence. Outdoor PM_2.5 levels were assumed to remain unchanged during each hour. The descriptive summary for hourly outdoor PM_2.5, indoor PM_2.5, RH, and temperature is shown in Appendix Table A1. The outdoor PM_2.5 levels ranged from 33 to 111 μg/m³ during the study, with a mean (standard deviation, SD) of 64 (17) μg/m³. This is more than five times higher than the typical mean outdoor PM_2.5 level (< 10 μg/m³) in this region during non-wildfire seasons (Huang et al., 2020; Xiang et al., 2020). With data pooled for all residences, the mean (SD) indoor PM_2.5 levels, when the PACs were absent and present, were 47 (24) μg/m³ and 14 (7) μg/m³, respectively. Indoor PM_2.5 levels increased significantly, up to ~250 μg/m³, when cooking events occurred. The impacts of cooking on indoor PM_2.5 lasted for several hours and varied in different residences, reflecting the difference in PM_2.5 total decay rates for different residences. Excluding the impacts of cooking events, indoor PM_2.5 levels were generally lower than outdoor levels. Additionally, the use of auto-mode PAC significantly reduced indoor PM_2.5 levels, excluding the impacts of cooking.

Fig. 2.

Time-series plots of one-minute indoor and outdoor PM_2.5 concentrations for each residence. Outdoor PM_2.5 levels were assumed to remain unchanged during each hour.

There was variation in the relative magnitudes of outdoor and indoor PM_2.5 levels for each residence. The different results among the residences reflect the variations in outdoor PM_2.5 levels, cooking events, and building characteristics over time and residences. Indoor RH and temperature are comparable between the filtration and non-filtration scenarios, with mean differences of 1.6% and 0.2 °C, respectively (Table A1).

3.3. I/O_INF and I/O_PAC

Based on the inclusion criteria described in the Method, Table 2 summarizes the residence-specific periods selected for calculations of I/O_INF and I/O_PAC. Accordingly, Fig. 3A and Appendix Table 2 show the residence-specific summary of PM_2.5 I/O_INF and I/O_PAC. Among the seven residences, the mean (SD) I/O_INF ranged from 0.33 (0.06) (Residence R1) to 0.76 (0.05) (Residence R6), with a mean (SD) of 0.56 (0.13). Generally, older buildings (R2, R3, R6), which were built before the 1960s, tend to have larger I/O_INF. However, building age is not the only factor that accounted for the I/O_INF differences among the residences. For instance, Residence R1 (built in 1988) had a lower mean I/O_INF than R4 and R5 (built in 2018–2019).

Table 2.

Residence-specific periods selected for calculations of I/O_INF and I/O_PAC.

Residence	I/OINF	I/OPAC
R1	09/16 14:00 – 09/17 11:00	09/17 14:00 – 09/17 20:00, 09/17 23:00 – 09/18 12:00
R2	09/16 15:00 – 09/16 16:00, 09/17 05:00 – 09/17 09:00	09/17 21:00 – 09/18 11:00
R3	09/16 17:00 – 09/17 11:00	09/17 20:00 – 09/18 13:00
R4	09/16 17:00 – 09/17 12:00	09/17 20:00 – 09/18 13:00
R5	09/17 14:00 – 09/17 17:50	09/18 11:00 – 09/18 15:00
R6	09/16 16:00 – 09/17 20:30, 09/18 01:00 – 09/18 13:00	-
R7	09/16 12:00 – 09/17 10:00	-

Fig. 3.

(A) Measured mean (standard deviation) of PM_2.5 I/O_INF and I/O_PAC for each residence; (B) mean (standard deviation) of the portable air cleaner (PAC) effectiveness. I/O_INF represents infiltration factor; I/O_PAC represents indoor/outdoor ratio with a portable air cleaner on.

Except for Residences R6 – R7, eligible periods were found to calculate I/O_PAC for the recruited residences. Among the five residences which utilized auto-mode PACs, the mean (SD) I/O_PAC ranged from 0.09 (0.02) (Residence R1) to 0.29 (0.05) (Residence R3), with a mean (SD) of 0.19 (0.09). Compared with non-filtration scenarios, PM_2.5 I/O ratios significantly reduced in all five residences. The effectiveness of the PAC, ε, was calculated based on Equation (7). As shown in Fig. 3B, the variation in ε, which ranged from 48% (11%) to 78% (4%), indicates the differences in building characteristics.

3.4. Decay-related parameters

According to the inclusion criteria described in the Method, Table 3 summarizes the residence-specific periods selected for calculations of k_TN and k_T. No eligible periods were available to estimate either k_TN for Residences R1, R3, R4, R7, or k_T for R2 and R4 – R7. Based on the measurements during the eligible periods, Table 4 presents the residence-specific means (SDs) of k_TN, k_T, AER, k_d, and k_PAC. k_TN and k_T, obtained from the exponential fitting of indoor PM_2.5 decay/increase curves, were significant (p < 0.05). The mean AER, or air infiltration rate, ranged from 0.30 (0.13) h⁻¹ to 1.41 (3.18) h⁻¹ among R1 – R6. The oldest residences, R3 (built in 1918) and R6 (built in 1906), had the largest AERs (> 1 h⁻¹). They were followed by R2 (built in 1948), with mean AERs of 0.53 (0.51) h⁻¹. The newest residences, R1 (built in 1988) and R5 (built in 2019), had the smallest AERs (~ 0.3 h⁻¹). The deposition rate, k_d, ranged from 0.10 (0.54) h⁻¹ to 0.49 (0.47) h⁻¹. R1 reported that a ceiling fan (no filter) was used during the experiment, partly resulting in the larger k_d. Relative to AER and k_d, k_PAC was 2–4 times larger, with means (SD) ranging from 1.50 (1.52) h⁻¹ to 2.12 (4.68) h⁻¹.

Table 3.

Residence-specific periods selected for calculations of PM_2.5 total decay rates (k_TN and k_T).

Residence	kTN	kT
R1	-	09/17 20:30 – 09/17 21:00
R2	09/16 19:00 – 09/17 05:00, 09/17 13:00 – 09/17 15:00	-
R3	-	09/17 16:30 – 09/17 17:00
R4	-	-
R5	09/17 18:30 – 09/17 23:00	-
R6	09/17 22:00 – 09/17 23:00	-
R7	-	-

Table 4.

Residence-specific means (standard deviations) of indoor PM_2.5 decay-related parameters.

Residence	kTN (h−1)	kT (h−1)	AER (h−1)	kd (h−1)	kPAC (h−1)
R1	-	2.90 (2.48)	0.32 (0.28)	0.49 (0.47)	2.10 (1.82)
R2	0.79 (0.75)	-	0.53 (0.51)	0.26 (0.28)	1.50 (1.52)
R3	-	3.90 (8.57)	1.41 (3.18)	0.36 (0.87)	2.12 (4.68)
R4	-	-	-	-	-
R5	0.45 (0.19)	-	0.30 (0.13)	0.15 (0.07)	1.64 (0.78)
R6	1.37 (6.69)	-	1.27 (6.22)	0.10 (0.54)	-
R7	-	-	-	-	-

4. Discussion

Despite the great public concerns on the high PM_2.5 exposure during wildfire seasons, few studies have evaluated PACs’ efficacy on reducing indoor PM_2.5 levels during such episodes in the US. To our knowledge, the present study is the first to examine the impacts of HEPA-based PAC use on indoor PM_2.5 concentrations during wildfire episodes in US residences, and the first to evaluate the efficacy of auto-mode PACs during wildfire episodes. The results of this study show that the mean (SD) PM_2.5 I/O ratios for the recruited residences, with the PAC absent and present, were 0.56 (0.13) and 0.19 (0.09), respectively. The PAC effectiveness results suggest that using a HEPA-based auto-mode PAC can significantly reduce indoor PM_2.5 levels by 48–78%, after adjusting for outdoor PM_2.5 levels and indoor emission sources. This study also estimates indoor PM_2.5 decay rates from air infiltration, deposition, and PAC removal by combining the steady-state and dynamic models.

This study utilized a detailed time-activity questionnaire (in minutes) and real-time indoor PM_2.5 measurements to precisely identify periods eligible for steady-state and dynamic models. As a result, the estimation of PM_2.5 I/O ratios and PAC effectiveness was well adjusted for impacts of indoor emission sources. Additionally, the algorithm developed in this study enabled us to reasonably estimate AERs and other decay-related parameters without any additional trace-gas measurements (e.g., CO₂, SF₆). Window-closed AERs within each residence were assumed as constants during the two-day experiments, potentially leading to biased results. However, the within-residence variation in window-closed AERs and PM_2.5 deposition rates should be marginal within such a short period (Pandian et al., 1998; Persily et al., 2010). Over the two days of this study, the predominant wind direction (mostly east/south) and wind speed (ranging from 1–2 m/s) were relatively stable (Washington State Department of Ecology, 2020). In our analysis, the AERs and other decay-related parameters were determined using the residence-specific data. In this way, the residence-specific characteristics were well adjusted. However, this approach should be used cautiously for long-term scenarios, especially when the status of windows and doors is unclear.

Compared with our previous study utilizing the same auto-mode PACs during non-wildfire scenarios (Huang et al., 2020), the effectiveness in the present study is significantly higher (48–78% versus ~30%). The difference in indoor PM_2.5 levels can explain the PAC effectiveness difference. The indoor PM_2.5 levels in this study without PAC use are much higher than those reported in Huang et al. (2020) (mean: 47 μg/m³ versus 14 μg/m³). As auto-mode PACs automatically adjusted the fan speed based on the surrounding PM_2.5 levels, higher indoor PM_2.5 levels resulted in larger PAC airflow and higher effectiveness. This study, combined with our previous research (Huang et al., 2020), confirms the efficacy of auto-mode PACs for multiple scenarios.

There is a growing interest in building do-it-yourself (DIY) box fan air cleaners by attaching an air filter to a box fan during a wildfire smoke event (US Environmental Protection Agency, 2019; Washington State Department of Health, 2020). Despite the concerns that the box fan motor may overheat when operated with a filter attached, such box fan air cleaners have been widely used in public during wildfire episodes. Nevertheless, the filtration efficacy of these DIY devices is currently unclear. In addition to the measurements made in the seven residences shown above (R1–R7), this study also examined the effectiveness of a DIY box fan air cleaner used in one Seattle residence (labeled R8 in the Appendix) during the wildfire episode. This residence underwent a 4-hour no filtration session, followed by a 3-hour filtration session and then a 12-hour no filtration session. The DIY box fan PAC, which consisted of a commercially-available box fan (Model B20200, Lasko, US) and a minimum efficiency reporting value (MERV) 13 filter (FilterBuy, Alabama, US), was used for the filtration. Without a filter, the manufacturer-specified flow rate for the box fan was 3000 m³/hour. Theoretically, the MERV 13 filter can remove ≥50% of PM_0.3–1.0 and ≥85% of PM_1.0–3.0 (ASHRAE., 2017). Based on the approach shown in the Methods section, the mean (SD) infiltration factor and effectiveness of this box fan PAC were 0.72 (0.05) and 59% (25%), respectively (see more details in the Appendix). Although more studies are warranted to examine the efficacy of different box fan PACs, the results support that some box fan PACs can result in meaningful reductions of indoor PM_2.5 levels during wildfire episodes.

In this study, windows and doors were kept closed for all residences over 1–2 days. However, some people may keep their windows open for a short period (e.g., 1 hour/day) to get some “fresh air” during a wildfire episode. In addition to the measurements mentioned above, one-hour PM_2.5 data with windows open and PAC off were collected in Residence R1. The PM_2.5 I/O ratio, absence of indoor sources, was approximately 0.92. Based on the steady-state model and the determined p and k_d in R1, the mean (SD) AER with windows open was 6.77 (6.30) h⁻¹. With such large AERs, it can be inferred that the PM_2.5 I/O ratio can generally be >0.7 even with the PAC on. Despite the variation in AERs, the fact remains that the PACs are not very effective with windows open. When opening windows is necessary (e.g., aiming to dilute indoor CO₂), it is meaningful to choose periods when the ambient PM_2.5 level is relatively low.

The I/O ratios reported herein were calculated based on outdoor PM_2.5 data from the nearest upwind governmental AQM station to each residence (2–10 km away). Appendix Fig. A3 shows the hourly variations of outdoor PM_2.5 levels during the study period at three governmental AQM stations in the Seattle region, 8–15 km apart from each other. The concentrations at all stations were similar, with pairwise mean absolute percentage differences (MAPD) of 12–14% between stations, suggesting that outdoor PM_2.5 levels measured at these AQM stations can reasonably represent those close to the residences in this study. Also, pollutant levels measured at governmental AQM stations have been commonly used to estimate indoor and personal exposure (Chen et al., 2017; Day et al., 2018; Day et al., 2017; Fisk and Chan, 2017b; Xiang et al., 2021; Xiang et al., 2019a; Xiang et al., 2019b). The station-based I/O ratios developed in this analysis can provide a reasonably accurate and convenient way to estimate indoor PM_2.5 concentrations using AQM station data for future studies.

Linear regression models are commonly used in time-series air pollution studies to examine the efficacy of intervention measures/policy after controlling for some factors (e.g., RH and temperature) (Holmes et al., 2020; Huang et al., 2020; Neumaier and Schneider, 2001; Xiang et al., 2020). In this study, indoor RH and temperature are comparable between the filtration and non-filtration scenarios, without much variation over the two days. Thus, by comparing the empirical data, the current approach is sufficient to explore the efficacy of PAC use. The results obtained from this study were based on measurements in seven residences from five dense communities in the Seattle region. More studies with larger sample sizes are warranted to investigate whether the results herein can be generalized to other residences in this region.

Wildfire events will continue and likely worsen in the US in the coming decades due to climate change (Abatzoglou and Williams, 2016; Littell et al., 2009; Parry et al., 2007). It is crucial to reduce wildfire-related PM_2.5 exposure and adverse health impacts during these events. Given that people spend most of their time indoors, a substantial reduction in indoor wildfire-related PM_2.5 exposure and attributed morbidity and mortality may be achieved using PACs during these episodes.

5. Conclusions

This study shows that the outdoor PM_2.5 levels were as high as > 100 μg/m³ during the wildfire episodes in the Seattle region. Without accounting for indoor emission sources and air filtration, the indoor PM_2.5 levels, with windows closed, were 33–76% of outdoor levels for the recruited Seattle residences. It suggests that staying indoors, a common protective measure during wildfire episodes, is insufficient to prevent people’s excess exposure to wildfire smoke. Using a HEPA-based portable air cleaner in auto mode can significantly reduce indoor PM_2.5 levels by 48–78%, after adjusting for outdoor PM_2.5 levels and indoor sources. Such analysis provides quantitative evidence to support the utilization of auto-mode PACs during wildfire events in the US.

Highlights.

• PM_2.5 infiltration factor was measured in seven residences during wildfire episodes.

• The efficacy of a portable air cleaner (PAC) was evaluated during wildfire episodes.

• Indoor PM_2.5 sources were adjusted using a minute-resolved time-activity diary.

• An algorithm was developed to separate indoor PM_2.5 decay from multiple pathways.

• Infiltration factor ranged from 0.3–0.8 while PAC effectiveness ranged from 48–78%.